Author(s):
Need: Recent studies of Teaching Assistant (TAs) interactions with students during one on one office hours suggest that TAs often fail to use good pedagogical practices and often struggle to debug erroneous students submissions. The goal of this project is to create a set of tools that will enable student to submit their erroneous code when they need help, generate a set of automatic corrections, and provide those corrections to a TA before they meet with a student. We refer to this set of tools as TEA, the Teaching Engagement AssistantGuiding Question: We have two primary guiding questions:1) What is the impact of providing automated corrections on the quality of the interactions between TAs and students during one on one office hours?2) What are the challenges that arise in an authentic usage context and how can they best be supported in the system?Outcomes: We have two kinds of outcomes: the creation of TEA and new knowledge about how TAs and students interact.We have created two systems. We have a help queue that allows students to submit their erroneous code, analyzes it and suggests corrections, and displays the list of students waiting for help to enable a team of TAs to manage a group of students needing help. We have also created a visualization that walks TAs through the corrections to student code and shows where the behavior of the code differs from a correct solution.We have also collected video recordings of TA student interactions and performed a qualitative analysis of their interactions. Our results suggest that there is a relationship between increased debugging time and effort and a decrease in the use of good pedagogical practices when interacting with students. We have also identified kinds of practices and debugging strategies that TAs use that can negatively and positively influence their interactions. Additionally, we have collected student submissions when seeking help and are currently analyzing these submissions to understand any patterns.Broader Impacts: The TEA system is potentially useful in courses that use Python and can be tested using unit tests. We are using the TEA system in the context of a Data Science course at the University of Michigan. And, we have extended the core system so that it works with a Python-based educational notebook and are piloting that use through a second course at Washington University. Through both of these courses, we have been tracking and responding to the real-world challenges that have arisen. We are beginning to develop TA training materials around the use of TEA as well as the good practices that we have identified through our qualitative study. We believe that both the system and the TA practices will be useful to other courses with a similar technical setup.
Coauthors
Caitlin Kelleher, Washington University, St Louis, MO; Barbara Ericson, University of Michigan, Ann Arbor, Michigan
